Why Learning To Read Is Easier In Welsh Than In English .

Applied Psycholinguistics 22:4Ellis & Hooper: Reading in Welsh and English574monosyllabic and eight disyllabic words in their language than were matchedFrench or English children (Goswami, Gombert, & De Barrera, 1998); andTurkish children are able to read and spell with high degrees of accuracy by theend of the first grade (Öney & Durgunoglu, 1997).Segmental phonological awarenessWhereas young children are aware of the structure of spoken language at thesyllable, onset, and rhyme levels, the representation of segmental information atthe phoneme level seems partly dependent on learning to read an alphabeticorthography. The written graphemes provide explicit feedback that clarifies thephonemic code (Brown & Ellis, 1994; Bradley & Bryant, 1983; Ellis & Large,1987; Goswami & Bryant, 1990; Morais, Alegria, & Content, 1987; Wimmer,Landerl, Linortner, & Hummer, 1991), and explicit focus on the representationsof sounds as they are written during spelling instruction seems to push alongthe acquisition of explicit phoneme segmentation abilities (Ellis & Cataldo,1990; Frith, 1985; Goswami & Bryant, 1990). This suggests that the representation of phonemic information as a consequence of the acquisition of readingshould depend on the degree of consistency or ambiguity of the system of symbol–sound mappings in a language. Research has demonstrated this to be so:phonemic segmentation develops in a rapid spurt in languages such as Greek orGerman, in which such graphemic feedback is a reliable index of segmentalphonology, and it tends to be more protracted in orthographies such as Frenchor English, which are more ambiguous in this respect (Cossu, Shankweiler, Liberman, Katz, & Tola, 1988; Frith, Wimmer, & Landerl, 1997; Goswami et al.,1998; Goswami, Porpodas, & Wheelwright, 1997; Wimmer & Goswami, 1994).Onset-rime awareness is a predictor of reading development for English, withits relatively high degree of spelling–sound consistency at the level of rimecompared to the level of the vowel phoneme, but not in German, in whichphoneme awareness is a much stronger predictor (Wimmer et al., 1991; Wimmer, Landerl, & Schneider, 1994).Reading strategyJust as fluent readers’ ability to read nonwords demonstrates the availability ofa synthetic route to reading in which the pronunciation is assembled on the basisof known symbol–sound associations, so their ability to pronounce irregular orinconsistent words implicates the availability of a direct reading route in whichthe word as a whole is used to access its pronunciation. Ab initio readers sometimes treat words like pictures and learn to relate the overall word shape to theword name. This first phase of logographic reading is a natural consequence ofthe use of flash cards and look-and-say methods. Skilled readers, on the otherhand, access pronunciation of the word as a whole on the basis of an analysisof the orthographic sequence of letters. Learners of English thus use differentbalances of predominance of different strategies of reading at different stagesof development: they move from logographic whole-word lookup, through analphabetic stage where phonology is assembled, to skilled reading, which ispredominantly orthographic whole-word lookup, at least for high-frequency

Applied Psycholinguistics 22:4Ellis & Hooper: Reading in Welsh and English575words (Frith, 1985; Marsh et al., 1981). Readers of transparent orthographies aremore likely to succeed in reading by means of alphabetic reading strategies thanreaders of oblique orthographies, and their differential histories of success or failure may well bias the reading strategy adopted by these learners, with readers oftransparent scripts being more likely to rely on alphabetic reading strategies.Again, there is support in the literature for this. Learners of German (Wimmer & Goswami, 1994) and Spanish (Lopez & Gonzalez, 1999) are more ableto read nonwords than are learners of English (Rack, Snowling, & Olson, 1992).Öney, Peter, and Katz (1997) showed that readers of Turkish are more likely touse synthetic phonology during word recognition, in that they were more affected by a spoken pseudoword prime that rhymed with the target words, thanwere American children – although, in support of stage theories of reading development, this disparity was greatest with beginning, second, and fifth gradereaders, and diminished in college students. Finally, adherence to an alphabeticdecoding strategy is likely to produce errors that are mispronunciations that donot sound like real words, whereas whole-word reading strategies are morelikely to generate erroneous real-word response errors. Accordingly, Wimmerand Hummer (1990) showed that the majority of German children’s readingerrors were nonwords, whereas young children reading English make frequent(wrong) real-word errors (Seymour & Elder, 1986; Stuart & Coltheart, 1988).Reading disorderOrthographically ambiguous languages such as English may be expected to posemuch more of a challenge in learning the symbol–sound pairings for childrenwho are impaired in phonological processing. Goswami (2000) argued that phonological deficits should be more associated with difficulties in learning to readopaque orthographies than transparent languages. Again, there is some evidencein support of this. Many children who are developmentally dyslexic in Englishhave phonological-processing deficits (Ellis, 1981; Frith, 1981; Snowling, 1998;Vellutino, 1979), to the degree that Frith characterized the disorder as “a failureof alphabetic skills” (Frith, 1985, p. 324). German dyslexic children, however,show much less marked difficulties in reading nonwords, a task that indexesphonemic recoding ability (Landerl, Wimmer, & Frith, 1997), although they doshow speed deficits. In contrast to English dyslexic children, Dutch dyslexicsshow no deficits in phonological awareness at the syllable and rhyme levels andtheir problems on phoneme deletion tasks do not persist into adulthood (DeGelder & Vroomen, 1991), whereas such difficulties do remain in adult Englishdyslexics (Bruck, 1992).Taken together, these findings suggest that orthographically transparent languages (a) promote faster rates of reading and spelling acquisition, (b) allowfaster development of phonemic awareness, (c) encourage an alphabetic readingstrategy, and (d) protect, to some extent, against phonological deficits as a causeof developmental dyslexia. However, there is one long-standing difficulty inthese cross-linguistic comparisons: they compare children who have learned toread different languages but who have also been taught by different teachers, indifferent classrooms, in different schools, using potentially different methods of

Applied Psycholinguistics 22:4Ellis & Hooper: Reading in Welsh and English576instruction, and in different cultures. It is hard to control all of these potentialconfounds.Britain affords an interesting contrast of written languages with the extremeopacity of English alongside the very transparent Welsh orthography, and theparticular milieux of North Wales may allow more control of these instructionaland cultural potential independent variables than has been possible in prior studies. The writing system of Welsh is so regular, with the mappings from graphemes to phonemes being so consistent and unambiguous, that Welsh dictionariesnever include pronunciation entries for the lexis (Ball & Jones, 1984; see Williams, 1994, for a full description of the text-to-speech synthesis rules). Phonemes can be represented with either single or double letter graphemes. Thereare 29 letters in the Welsh alphabet, including several letter combinations (ch,dd, ff, ng, ll, ph, rh, th), which each represent one sound. As in English, vowels(a, e, i, o, u, w, y) can be either long or short. As shown in Appendix A, withconsonant graphemes there is almost a one-to-one mapping between graphemeand phoneme, even the consonant digraphs are invariant (e.g., ff for /f/). Thevowels are a little more problematic. Orthographic y may be realized as a schwain nonfinal syllables of polysyllables; it may also represent the first nonvocalicpart of a diphthong (yw, /iu/), or the second consonantal part of the diphthong(wy, /ui/). Welsh also permits epenthetic vowels, these being vowels that arepronounced but are not shown in the orthography. The graphemes i and w arealso variable – i can be either a long or short vowel (/ii/, /i/), or the palatalglide /j/. The grapheme w is the hardest to realize, with four or five possibleinterpretations. Notwithstanding these few irregularities (this list is reasonablyexhaustive), Welsh is a highly transparent orthography in comparison to English, in which Treiman et al. (1995) estimated that the pronunciation of vowelsis only 51% consistent over different words; consider, for example, the a in cat,call, car, cake, and care.In parts of North Wales, the two languages are spoken and read side by side.In particular, in the North East region, parents choose whether they want theirchild to attend English medium or Welsh medium schooling. Generally, it ismonolingual English parents who choose the former and bilingual Welsh/English parents who speak Welsh in the home who choose the latter. Theseschools serve the same geographical catchment area, are administered by thesame local education authority, and follow similar curricula and teaching approaches; the only real difference is the language of instruction. Children sentto Welsh medium primary schools are taught to read in Welsh, and they areonly introduced to reading in English in Year 3, usually at 8 years old.This situation permitted a profitable replication and extension of the workdone on orthographic transparency to date. In the first place, it allowed for theassessment of effects of script in readers who are matched for geographicalarea and concomitant socioeconomic variation, broad instructional milieu, andeducational background, factors that are not as tightly controlled in many of thestudies reviewed here. In the second place, we wished to do a better job ofassessing reading progress than that allowed by the use of small language samples (sometimes only 8 or 16 words) or the translation equivalent pairs that arecommon in the studies described here; we wanted to know just how far through

Applied Psycholinguistics 22:4Ellis & Hooper: Reading in Welsh and English577the language English and Welsh readers had gone after similar time or task.Finally, we wanted to identify whether, despite their very similar methods ofreading instruction, Welsh and English children adopt different strategies ofreading as a consequence of the different degrees of transparency of these written orthographies.It is difficult to compare the first language (L1) acquisition of literacy indifferent languages because the stimuli must necessarily be different. Some ofthe reviewed studies investigated whether translation equivalents are acquiredequally easily in the different languages, but there is no guarantee that translation pairs are equally frequent in their respective languages. Others comparedthe ease with which non-words are read in the different languages, but howeverwell this gives a measure of the generalizability of the child’s reading skill, bytheir very definition, non-words are not items of the languages themselves. Nor,of course, is comparison of performance on standardized reading tests in different languages going to avail, because the standardizations are made with reference to large samples of readers of one language only. Instead, the answer tothe question, Is it easier to learn to read in language X rather than language Y?lies in the use of well-matched reading tests in the two languages. But for whatshould they be matched? The tests must be carefully constructed so that theitems are equally representative of language X and of language Y, and childrenof the same age must have had equal opportunity to experience the parallelitems. Sampling theory informs us that representative samples must be randomlyselected, although their accuracy can be increased by stratifying on importantpotential independent variables. The relevant independent variable with regardto experience is frequency. Thus, the tests should be constructed as randomselections of each language’s lexis that are stratified by frequency, with pairs oftest items in the two languages being yoked in this regard. However, the matching process must not control any other factors, such as word length, imageability, utility, morphological complexity, syntactic role, semantic richness, orthographic complexity, sound–symbol consistency, or any other intrinsic aspect ofthe languages under study or their learnability. Everything to do with learningopportunity should be matched; everything to do with language should be freedto vary. Kempe and MacWhinney (1996) explained this principle and illustratedits use in the construction of matched lexical decision tests based on log frequency stratified samples from complete frequency counts of the two languagesof concern in their assessment of cross-linguistic vocabulary knowledge. Therationale is based simply on input-driven perspectives of language acquisition:The learnability of an item is largely dependent upon the amount of experiencea learner has of it and of its kind. Thus, if children of languages X and Y startlearning to read at the same time and have roughly equivalent time-on-task,then, if the two orthographies are equally difficult to acquire, learners of similarexperience should be able to read down to roughly to the same frequency levelof sample test item in the two languages.This study was therefore designed to compare the rate of acquisition of orthographically transparent Welsh and orthographically opaque English using reading tests that were equated for amount of exposure. The word types that composed million token word frequency profiles for English and Welsh were sorted

Applied Psycholinguistics 22:4Ellis & Hooper: Reading in Welsh and English578in decreasing frequency of occurrence and sampled so that a test word wasselected that matched (roughly) every decreasing step of 10,000 word tokens.The two lists of 100 words were then used as frequency-matched test samples ofwritten English and Welsh. The study investigated (a) whether Welsh childrenprogressed further in reading than English children of similar reading experience, (b) whether they differed equally in reading comprehension abilities, (c)whether their reading latencies were indicative of different reading strategies,and (d) whether the two groups of learners made qualitatively different patternsof error that also indexed a difference in reading strategy.METHODSchoolsSix primary schools in the Wrexham area of northeast Wales agreed to participate in the study. Three of these schools were Welsh speaking and three wereEnglish speaking. They were matched on a number of variables, including theircatchment areas, classroom sizes, and teaching methods, as determined by detailed interviews with the teachers.Reading instruction in the English schools was done broadly as follows. Innursery school, children were taught the shapes of most letters of the alphabetand introduced to book use. They were read to on a daily basis, and there wasemphasis on repetition and rhyming in the reading books used. In the “Reception” class, Big Books (Oxford Reading Tree; Oxford University Press) wereused by all schools, with approximately 30 min per day dedicated solely toreading. These large-sized books were used when reading aloud to a wholeclass, with children being able to see both the written texts and the large illustrations in order to develop individual, group, and whole class reading and listeningskills. Flash cards were also used to introduce individual words and to formsimple sentences. Children were encouraged to take books home to read. Therewas much use of word building, pattern recognition, and odd-one-out games. InYear 1, the Oxford Reading Tree scheme was incorporated as a reading tool ineach school and used to implement the literacy hour. Rhyme and analogy materials were used to develop phonological awareness, poetry was introduced toencourage children’s awareness of rhythm and rhyme, and Jolly phonics (JollyLearning Ltd.) was also used to helping them to develop phonological skills.Year 2 saw the final stages of the Oxford Reading Tree and some use of theFirst Steps Scheme (Heinemann). Approximately 30 min was spent each day onreading, mainly within groups, but children were also assessed on a one-to-onebasis with their teachers. They were encouraged further to read two or threebooks a week at home, with their parents or guardians being asked to commenton reading development.Reading instruction in the Welsh schools was much the same, in that it usedthe same materials in translation, was guided by the same curriculum, and devoted the same amount of time to the activities. The alphabet and letter shapeswere introduced in nursery school. Big books were used in Reception. In Years1 and 2, children had approximately 30 min of reading daily, with opportunities

Applied Psycholinguistics 22:4Ellis & Hooper: Reading in Welsh and English579for individual, group, and class reading. The Oxford Reading Tree scheme alsoprovided the core curriculum materials in the Welsh schools, with stages 1through 5 being a direct translation of the English scheme, although no translations are given for the rhyme and analogy resources provided for English-speaking children. A reading corner was available in each classroom. A phoneticteaching approach was a large part of Year 1 instruction. In terms of homework,books were taken home every night, and children were expected to read at leastthree times a week. Diaries were used by parents and teachers to keep track ofreading development. Children were also encouraged to search for informationin different types of literature on particular themes which interested them.ParticipantsA consent form, a brief description of the study, and a questionnaire concerninglanguage use in the child’s home and their prior educational history was sent toevery pupil in Year 2 in the six participating schools. The questionnaire, basedon Lyon (1996), included five questions concerning the language that wasmostly used when the child was in conversation with their parents or guardians,siblings, grandparents, and friends. On the basis of these returns, 20 Welshchildren were selected from the Welsh primary schools and 20 English childrenfrom the English schools. There were 17 girls and 23 boys. They were either 6or 7 years old and attended Year 2. The majority of the Welsh children spokeWelsh with their immediate family, although a few only spoke Welsh withextended family who lived in the child’s home and with friends. The Englishchildren were monolingual. Children were selected to participate only if theyhad been exposed to reading in their L1 and if they had normal academic levelsof attainment. The two groups were matched with regard their participants’ exposure to reading, with the majority having only been introduced to reading atschool; however, six Welsh and seven English children had received some basicprereading instruction at home prior to school entry.English and Welsh participants were further matched on their academic levelsusing the results from the Key Stage 1 maths exams taken in May 1999 in Year2. These exams were translation equivalents of the same problems, with possiblegrades ranging from Level 1, through Levels 2a, 2b, and 2c, to Level 3. Thetests included workbooks in which each child was required to read and answerquestions individually. When the maths grades were converted to a 5-pointscale, the Welsh group (M 3.55, SD 1.00) and English Group (M 3.35,SD 1.04) did not differ significantly, t(38) 0.62, ns. We take this equality toindicate that the two groups of children were broadly the same in terms of nativeability and quality of schooling.MaterialsThe English and Welsh reading tests were each compiled by sampling wordsfrom 100 successive strata of decreasing written word frequency in the language. This ensured that both lists were closely matched on frequency levels.Word form lists were used rather than lemmas. The English test was formed by

Applied Psycholinguistics 22:4Ellis & Hooper: Reading in Welsh an

and exemplar-based (Ellis & Hicks, 2000) computational models. These investigations have compared the learnability and processing of words of different degrees of spelling–sound ambiguity within a language. What about the large cross-linguistic issue: what are the effects of the overall ambiguity of